dataset_path: pangyyyyy/STARE

generation_kwargs:
  max_new_tokens: 8192
  temperature: 0.0
  top_p: 1.0
  num_beams: 1
  do_sample: false

output_type: generate_until
doc_to_visual: !function utils.stare_doc_to_visual
doc_to_text: !function utils.stare_doc_to_text
doc_to_target: utils.stare_doc_to_target
process_results: !function utils.stare_process_results

metric_list:
  - metric: stare_score
    aggregation: !function utils.stare_aggregate_results
    higher_is_better: true

dataset_kwargs:
  token: True
  cache_dir: STARE
  force_download: true

metadata:
  strategy: CoT # ['Direct', 'CoT']
  use_lmms_judge: False